OR I G INA L ART I C L E

Anatomical Abnormalities in Autism?Shlomi Haar1, Sigal Berman3, Marlene Behrmann4, and Ilan Dinstein1,2

1Department of Brain and Cognitive Sciences, 2Department of Psychology, 3Department of Industrial Engineeringand Management, Ben Gurion University of the Negev, Beer Sheva 84105, Israel, and 4Department of Psychology,Carnegie Mellon University, Pittsburgh, PA 15213, USA

Address correspondence to Dr Ilan Dinstein. Email: dinshi@bgu.ac.il

AbstractSubstantial controversy exists regarding the presence and significance of anatomical abnormalities in autism spectrumdisorders (ASD). The release of the Autism Brain Imaging Data Exchange (∼1000 participants, age 6–65 years) offers anunprecedented opportunity to conduct large-scale comparisons of anatomical MRI scans across groups and to resolve many ofthe outstanding questions. Comprehensive univariate analyses using volumetric, thickness, and surface areameasures of over180 anatomically defined brain areas, revealed significantly larger ventricular volumes, smaller corpus callosum volume(central segment only), and several cortical areas with increased thickness in the ASD group. Previously reported anatomicalabnormalities in ASD including larger intracranial volumes, smaller cerebellar volumes, and larger amygdala volumeswere notsubstantiated by the current study. In addition, multivariate classification analyses yielded modest decoding accuracies ofindividuals’ group identity (<60%), suggesting that the examined anatomical measures are of limited diagnostic utility for ASD.While anatomical abnormalities may be present in distinct subgroups of ASD individuals, the current findings show that manypreviously reported anatomical measures are likely to be of low clinical and scientific significance for understanding ASDneuropathology as a whole in individuals 6–35 years old.

Key words: anatomy, autism, MRI, thickness, volume

IntroductionConsiderable effort has been devoted to the identification ofanatomical abnormalities in individuals with autism spectrumdisorder (ASD) (Courchesne et al. 2007; Amaral et al. 2008). Inthe current study,we analyzed anatomical data acquired from in-dividuals older than 6 years of age for whomMRI scans are avail-able in the Autism Brain Imaging Data Exchange (ABIDE)database. Previous studies of individuals in this age range havereported that, in comparison to controls, ASD individuals exhibitnumerous abnormalities including larger gray matter (Lotspeichet al. 2004; Hazlett et al. 2006; Ecker et al. 2013), white matter (Ha-zlett et al. 2006), amygdala (Bellani et al. 2013a), and hippocam-pus (Groen et al. 2010) volumes, smaller cerebellum (Scott et al.2009; Fatemi et al. 2012) and corpus callosum (CC; Bellani et al.2013b) volumes, and abnormal cortical thickness (Raznahanet al. 2010; Wallace et al. 2010). These findings have beeninterpreted as supporting evidence for different theories ofASD including, for example, the “amygdala theory of autism”

(Baron-Cohen et al. 2000) and the “underconnectivity” theory ofASD (Just et al. 2007). These findings, however, have not beenreplicated consistently in the literature and heterogeneousresults across studies with small samples of participants havedemonstrated a critical need for analyzing larger cohorts(Amaral et al. 2008).

More recent anatomical studies have also utilized multivari-ate classification techniques to identify patterns of anatomical

measures that differ across ASD and control individuals instead

of focusing on just one measure at a time. These studies have

reported remarkable accuracies in decoding the group identity

of single subjects (above 85%) when utilizing measures of

cortical thickness, geometry, curvature, and/or surface area

(Ecker, Marquand et al. 2010; Jiao et al. 2010; Uddin et al. 2011),

thereby implicitly suggesting that anatomical measures may

have clinical diagnostic value for ASD. While these initial results

seem promising, it is important to note that previous studies

sampled data from only 20–30 subjects in each group and

© The Author 2014. Published by Oxford University Press. All rights reserved. For Permissions, please e-mail: journals.permissions@oup.com

Cerebral Cortex, 2014, 1–13

doi: 10.1093/cercor/bhu242ORIGINAL ARTICLE

1

Cerebral Cortex Advance Access published October 14, 2014 at A

cquisitions DeptH

unt Library on O

ctober 15, 2014http://cercor.oxfordjournals.org/

Dow

nloaded from

reported successful classification with entirely different sets ofanatomical measures. The lack of consistency across studiesraises critical questions regarding the generalizability and repro-ducibility of findings to larger cohorts.

The release of theABIDE data exchange (DiMartino et al. 2014)offers a unique opportunity to perform more definitive anatom-ical comparisons by analyzing aggregated MRI data, collected in20 different international studies, from ∼1000 ASD and controlparticipants, ages 6–65 years. This dataset, which is an order ofmagnitude larger than any anatomical dataset published todate, offers substantial statistical power and the ability to dissoci-ate anatomical variability related to age, IQ, and scanning sitefrom true anatomical differences across control and autismgroups. As such, this dataset offers an unprecedented opportun-ity to determine the existence (or lack thereof ) of anatomicalabnormalities in children, adolescents, and adults with ASD.

Materials and MethodsThe data analyzed in the current studyare part of theABIDE data-base (http://fcon_1000.projects.nitrc.org/indi/abide). All data arefully anonymized as required by HIPAA regulations and all par-ticipating sites received local Institutional Review Board approvalfor acquisition of the contributed data.

Subjects

The ABIDE database contains aggregated MRI scans of 539 ASDindividuals and 573 typical controls aged 6–65 years who werescanned as part of 20 international studies. We performedanalyses on 2 samples from the database.

Strict SampleHerewe excluded ASD subjects older than 35 (21 subjects) and allfemales (65 subjects) given their low preponderance. Individualswithout IQ scores (34 subjects), those with low-quality MRI scans(26 subjects, see below), and those with gross anatomicalvolumes that exceeded 3 standard deviations from the mean ofthe site (8 ASD and 8 control subjects in total) were also excluded.The remaining 401 ASD subjects werematched to controls withinthe same site based on age (±5 years) and IQ (±10 points). ASDsubjects for whom there was no matched control were excluded.Final analyses were performed with 295 ASD and 295 controlsubjects (Table 1).

Relaxed SampleHere we excluded subjects with low-quality MRI scans and equa-ted the number of ASD and control subjects within each site byselecting control subjects that were closest in age to the ASD sub-jects. Left over subjects from sites with an unequal number ofASD and control subjects were excluded. Final analyseswere per-formed with 453 ASD and 453 control subjects (SupplementaryTable 1). In this sample subjects were NOTmatched with respectto gender, IQ, or age.

MRI Scans

All MRI data were acquired using 3-Tesla scanners with T1

weighted scans (1 × 1 × 1-mm resolution). The specific scanningparameters of each sample are available at http://fcon_1000.projects.nitrc.org/indi/abide.

Brain Segmentation

Cortical reconstruction and volumetric segmentation was per-formed with Freesurfer (http://surfer.nmr.mgh.harvard.edu/).The technical details of these procedures are described else-where (Fischl 2012). Briefly, this processing includes removal ofnonbrain tissue and segmentation of subcortical and corticalgray and white matters based on image intensity. Low-qualityMRI scans where the segmentation procedures failed or showedgeometric inaccuracies were excluded (26 ASD subjects).

ROI Parcellation

Parcellation was also performed using Freesurfer. Individualbrains were registered to a spherical atlas which utilized individ-ual cortical folding patterns to match brain geometry across sub-jects. Each brain was then parcellated into 148 graymatter and 32subcortical ROIs using the Destrieux anatomical atlas (Destrieuxet al. 2010). Finally, ROI labels were transformed back into eachsubject’s native space to compute the volume of each ROI.

Cortical Thickness and Surface Area

Both measures were computed using Freesurfer by estimatingthe gray/white matter boundary and the pial surface (Fischl andDale 2000). Cortical thickness was computed for each triangularvertex as the distance from one surface to the other. The numberof vertices varied across subjects according to individual corticalsurface area. ROI thickness was computed by averaging the verti-ces belonging to each ROI in each subject’s native anatomicalspace. Cortical surface area was computed per ROI using 2Dflattened representation of the cortical surface.

Regressing Out Site Age and IQ

Whenperforming themixed-model analysis using group and siteas main factors and age and IQ as covariates, we found that site,age, and IQ had significant effects onmost anatomical measures.To eliminate differences due to these confounding factors we re-gressed out the effects of site, age, and IQ from each of the ROIsfor each of the anatomicalmeasures separately. Thiswas done byperforming amultiple regression analysis of the anatomical data(including both ASD and control subjects) with the 3 variables(site, age, and IQ) and extracting the residuals. Since anatomicalmeasuresmay change in a linear (e.g., graymatter volume), expo-nential (e.g., white matter volume), or quadratic (e.g., atrophy atolder ages) manner with age, we included a linear, an exponen-tial, and a quadratic predictor for age in the multiple regressionanalysis. This analysis was performed for each ROI separately.

Classification

We used linear and quadratic (nonlinear) discriminant analysesimplemented in MATLAB to classify ASD and control subjectsaccording to anatomical measures. Discriminant analysis is asupervised multivariate classification method where each sub-ject is represented by a vector in a high dimensional space de-fined by the number of selected features (anatomical ROIs).Training data were classified into 2 classes (ASD and controls)by identifying a multidimensional hyperplane in the linear caseand a multidimensional quadratic surface in the nonlinear casethat optimally separates the 2 classes. The accuracy of the hyper-plane/surfacewas tested by assessing its ability to separate inde-pendent “testing data” that was not included in the “trainingdata” (see cross-validation below). Classification was applied

2 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

separately to five different anatomical feature sets: all ROIvolumes (180 features), subcortical ROI volumes (32 features),cortical ROI volumes (148 features), thicknesses (148 features),and surface areas (148 features).

10-Fold Cross-validation

The data were randomly split into 10 even-sized subsets, whichincluded an even number of subjects from ASD and controlgroups. Nine subsets (90% of the data) were used to train the clas-sifier and the remaining subset (10%) was used to test classifieraccuracy by assessing the proportion of subjects that were accur-ately decoded. This process was iterated 10 times such that eachof the subsets was used once for “testing” and the decodingaccuracies were averaged across iterations.

Leave-Two-Out Cross-validation

Here we used the same procedure as the “10-fold” validation butleft out 2 exemplars (one with ASD and one control) instead of10% of the data from each group for “testing” and the numberof iterations equaled n/2.

Randomization Test

We used a randomization test to determine statistical signifi-cance in the different analyses. In the univariate analyses, werandomly shuffled the ASD and control labels across subjectsand computed the difference in the particular measure (e.g.,intracranial volume) across groups. This process was repeated10 000 times to generate 10 000 difference values that representeda distribution of differences expected by chance (null distribu-tion). For the true (un-shuffled) value to be considered significant,it had to surpass the 2.5th or 97.5th percentile of the null distribu-tion (i.e., the equivalent of a P < 0.05 value in a two-tailed t-test).A separate randomization analysis was performed for each

anatomical measure. In the multivariate analyses, we similarlyshuffled the ASD and control labels and then performed identicalclassification and cross-validation procedures to those describedabove. This process was repeated 100 times to generate 100decoding accuracy values that represented a distribution ofaccuracies expected by chance (null distribution). A separaterandomization analysis was performed for each classificationanalysis with each set of anatomical features. For the real decod-ing accuracy to be considered statistically different from chancelevels, it had to exceed the 95th percentile of the relevant nulldistribution.

Multiple Comparisons

We used false discovery rate (FDR) correction (Benjamini andHechtlinger 2014) to control for the multiple comparisons prob-lem when assessing anatomical differences across multipleROIs (e.g., Fig. 2) . This correction ismore relaxed than the Bonfer-roni method, andwe used it in order to increase the sensitivity ofthe analyses so as not to miss any potential differences acrossgroups.

ResultsWhile the ABIDE database contains data from ∼1000 individuals,behavioral scores (i.e., IQ, ADOS, ADI) were not available for allsubjects and age and gender were not sufficiently matched with-in all sites. We, therefore, decided to perform the majority of ouranalyses with a subset of 590 participants (half controls) whowere selected using the following stringent criteria. Since 98%of the subjects were under the age of 35 and 85% were male, weselected only males 6–35 years old. An equal number of ASDand control subjects were sampled from each site and eachASD subject was matched with a control subject on IQ and agewithin the site. This selection ensured that there were no majorgroup differences on multiple demographic, biographic, and

Table 1 Strict sample phenotypic information

Site Scanner n Age IQ ADOS

ASD TD ASD TD ASD TD ASD

Mean SD Mean SD Mean SD Mean SD Mean SD

CALTECH SIEMENS Trio 5 5 22.6 4.2 22.1 3.7 113 11.1 116 6.1 10.8 2.7CMU SIEMENS Verio 8 8 24.9 4.1 25.8 4.7 110 11.6 109 6.1 13.0 3.2KKI Philips Achieva 10 10 10.4 1.4 10.4 1.4 107 13.3 111 11.4 10.6 1.8Leuven 1 Philips INTERA 12 12 20.4 1.9 22.7 2.6 110 12.5 113 10.5Max Munich SIEMENS Verio 13 13 19.9 9.2 20.1 8.9 109 9.7 109 8.1 9.8 3.9NYU SIEMENS Allegra 56 56 13.6 5.9 14.3 4.8 110 16.5 112 13.2 11.3 4.1OHSU SIEMENS Trio 8 8 11.1 2.1 10.0 1.1 117 12.6 115 10.1 8.6 3.7OLIN SIEMENS Allegra 13 13 16.2 3.0 17.4 3.4 114 16.2 114 17.1 14.2 3.7PITT SIEMENS Allegra 18 18 18.3 6.0 17.4 5.3 109 13.7 110 9.4 12.8 3.0SDSU GE MR750 8 8 14.6 1.7 14.3 1.3 113 9.5 113 7.5 10.6 4.1Stanford GE Signa 9 9 9.5 1.6 9.4 1.3 113 13.1 114 11.7 11.9 3.8Trinity Philips Achieva 22 22 16.7 3.0 16.4 3.1 111 13.3 111 12.4 10.6 2.9UCLA 1 SIEMENS Trio 27 27 13.2 2.6 13.3 2.2 104 13.0 105 10.4 10.5 3.3UCLA 2 SIEMENS Trio 5 5 12.3 1.9 12.2 1.1 105 9.4 108 8.8 13.5 3.1UM 1 GE Signa 26 26 12.8 2.4 13.1 3.2 105 15.9 108 9.8UM 2 GE Signa 11 11 14.6 1.6 15.6 1.7 114 13.0 112 9.1USM SIEMENS Trio 30 30 21.5 5.7 21.1 5.9 107 13.9 110 10.3 12.9 2.7Yale SIEMENS Trio 14 14 12.6 3.0 12.2 2.8 99 18.4 100 16.5

Number of subjects, mean and standard deviation of age, IQ scores, and ADOS scores (total communication and social scores from ADOS modules 3 and 4 only) are

presented for each site along with the type of MRI scanner used at the site. Note that ADOS testing was not performed at 4 of the sites (63 subjects with ASD).

Anatomical Abnormalities in ASD Haar et al. | 3

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

environmental variables that may have varied across inter-national sites. In addition, we repeated the analyses with largersets of subjects while using more relaxed sampling criteria andfound equivalent results to those described below. To demon-strate this, we present results from a larger sample containing906 participants who were not matched with respect to age,gender, or IQ (Supplementary Figs 6 and 7).

Intracranial Volume

To determine whether there were significant differences inintracranial volumes across the ASD and control groups, we per-formed a mixed-model analysis (MANCOVA) with 2 main factors(group and site) and 2 covariates (age and IQ). This analysisrevealed significant differences across groups (P = 0.04) and sites(P < 0.001), which co-varied significantly with age and IQ(P < 0.002), and a significant interaction between site and group(P = 0.02). Since the interaction between group and site was sig-nificant, we performed an additional analysis per site and as-sessed whether differences between groups were consistentacross sites (Fig. 1A). The results showed that significant differ-ences across groups were found in only 2 of the 18 sites, whichcontained only 26 of the 590 (4.4%) subjects. Removing thesesites from the analysis eliminated the significant differenceacross groups and the significant interaction between groupand site (P > 0.1 for both) thereby demonstrating that initialsignificance was driven by the small subset of subjects fromthese 2 sites.

The results of the mixed-model analysis demonstrated thatintracranial volumes varied significantly across sites, ages, andIQs. Since these are confounding factors in comparisons of

anatomy across control and ASD groups, we reanalyzed thedata after regressing out the variability associated with each fac-tor. This normalization procedure eliminated mean volume dif-ferences across sites, ages, and IQ levels, but did not affect thedifferences between autism and control groups within eachsite. Hence the 2 sites with significant intracranial volume differ-ences across groups before normalization also exhibit significantdifferences after normalization (Fig. 1D and SupplementaryFig. 1). Performing an independent t-test or randomization teston the normalized data showed that therewere no significant dif-ferences across ASD and control groups (P > 0.5, independent t-test and randomization test, Fig. 1B,E) and demonstrated thatthe negative finding was not due to variability associated withthese confounding variables.

In a final analysis, we also stratified the ASD individuals intosubgroups with weak (ADOS <10), moderate (ADOS = 10–14), andsevere (ADOS >14) symptoms asmeasured by the total social andcommunication ADOS scores. Both t-tests and randomizationtests did not reveal any significant differences between each ofthe three ASD severity groups and their matched control groups(P > 0.25 for all comparisons, Fig. 1C,F).

Gross Anatomy and ROIs of Special Interest

Equivalentmixed-model analyses were performed separately foreach of the following measures: white matter volume, gray mat-ter volume, cortical thickness, and cortical surface area as well ascerebellar graymatter, cerebellar whitematter, amygdala, hippo-campus, CC, and ventricular volumes. All volumetric measureswere first translated to percentages of intracranial volume inorder to compensate for differences in intracranial volumes

Figure 1. Scatter plot of intracranial volumes of individuals with ASD (gray) and controls (black) in each site (A). Sites are ordered according to age. Mean (SD) age and

scanner type are noted for each site (GE, General Electric; P, Phillips; S, Siemens). Bars represent the mean intracranial volume of each group across sites (B). An

additional scatter plot (C) represents intracranial volumes after separating ASD subjects into subgroups with ADOS <10, 10–14, and >14. Panels on bottom row show

equivalent analyses after regressing out variability associated with site, age, and IQ (D–F). Asterisks: significant differences (P < 0.05, randomization test, FDR corrected

for multiple comparisons). Horizontal black lines: mean across site (A, D) or group (C, F). Error bars: standard error of the mean across subjects.

4 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

across subjects. Here too, several measures exhibited significantdifferences across groups, but also significant interactions be-tween group and site. We, therefore, normalized differencesacross sites, ages, and IQ levels and re-examined group differ-ences using independent t-tests and randomization tests (Fig. 2).

These analyses revealed that only ventricular volume wassignificantly larger in the autism group when compared withcontrols (P < 0.001, randomization test, FDR corrected). All otheranatomical measures did not differ significantly across groups(P > 0.1). A “per-site” analysis showed that the vast majority ofsites did not exhibit significant differences across groups forany of the measures. In the case of cortical gray matter, 2 sitesexhibited contradictory findings. Effect sizes across groups weresmall with the largest one evident in the comparison of ventricu-lar volumes (d = 0.34).

Stratifying the ASD individuals into subgroups with weak(ADOS <10), moderate (ADOS = 10–14), and severe (ADOS >14)symptoms did not reveal any additional differences acrossgroups. None of the autism severity groups showed significantdifferences from the control group for anyof the anatomicalmea-sures including the ventricular volumes (P > 0.07, randomizationtest, FDR corrected, Supplementary Fig. 2). This shows that thedifference in ventricular volumes across groups is relativelyweak and does not appear reliably when sampling smallersubgroups of autism and control subjects.

Finally, we also examined whether there were volumetricgroup differences in specific segments of the CC. The CC ofeach subject was segmented into 5 equal parts along the anter-ior–posterior axis and the volume of each part was calculated(Fig. 3). The ASD group exhibited significantly smaller volumesin the central CC segment when compared with the controlgroup (P < 0.05, randomization test, FDR corrected). Note thatthe effect size of this finding was small (d = 0.2) and that signifi-cant differences appeared in only 2 of the 18 sites.

Additional Cortical and Subcortical Areas

Univariate comparisons across groups were performed for eachof 148 cortical and 32 subcortical ROIs that were defined in anautomated manner for each subject using the Freesurfer atlas(see Materials and Methods). Volumetric measures were normal-ized by total intracranial volume and variability associated withsite, age, and IQ was regressed out for each measure separately.Analysis of the cortical ROIs revealed that individuals with ASDexhibited significantly thicker cortex than controls in severalROIs including the right and left occipital poles, left middle oc-cipital sulcus, left occipital-temporal sulcus, left cuneus gyrus,right and left subparietal sulci, and left superior temporal gyrusand sulcus (P < 0.05, independent t-test, FDR corrected, Fig. 4).There were no significant differences in volumetric or surfacearea measures between groups. There were also no significantvolumetric differences between groups in any of the subcorticalROIs.

Correlations with IQ and ADOS

Intracranial, gray, and white matter volumes as well as corticalsurface were positively and significantly correlated with IQ inboth control and ASD groups (P < 0.05, randomization test, uncor-rected to increase sensitivity, Fig. 5, top andmiddle rows). Indivi-duals with ASD also exhibited significant positive correlationsbetween gray and white matter volumes and cortical surfacearea measures and ADOS scores (Fig. 5, bottom row). Note that

all correlations were computed after regressing out site and ageas performed in the analyses described above.

Classification

We used a linear discriminant analysis (LDA) classifier to decodethe group identity of individual subjects (Fig. 6). A 10-fold valid-ation scheme was used to assess the decoding accuracy of theclassifier such that 90% of the data were used to train the classi-fier and the left-out 10% was used to assess decoding accuracy.This procedure was performed 10 times, each time leaving outa different subset of subjects for testing. A randomizationanalysis was used to determinewhether decoding accuracies ex-ceeded chance levels (see Materials and Methods). Classificationwas performed using 5 anatomical feature sets: all ROI volumes(180 ROIs), only gray matter ROI volumes (148 ROIs), only subcor-tical ROI volumes (32 ROIs), cortical thickness (148 ROIs), andcortical surface area (148 ROIs).

Above-chance decoding accuracieswere foundwhen classify-ing all ASD and control subjects using subcortical volumes andcortical thickness measures (accuracies = 56% and 60%, respect-ively, P < 0.05, uncorrected to increase sensitivity). No other ana-tomical measures yielded significant, above-chance decodingaccuracies. When performing the same analysis separately onASD subgroups with different symptom severities, above-chancedecoding accuracies were found only for the ASD subgroup withsevere symptoms using subcortical volumes (accuracy = 60%,P < 0.05, uncorrected).

Additional analyses using a “leave-two-out” validationscheme with an LDA classifier or a 10-fold validation schemewith a quadratic discriminant analysis classifier (i.e., nonlinearclassifier) revealed similar decoding accuracies that did notexceeded 60% and exceeded chance level only with subcorticalvolumes and cortical thicknessmeasures (Supplementary Fig. 3).

Finally, we performed a per-site classification analysis withinthe 5 sites that contained at least 20 subjects in each group (Sup-plementary Fig. 4). This analysis revealed variable and weak de-coding accuracies across sites, which did not exceed 60% for anyof the sites when using any of the anatomical measures. Thissuggests that the poor decoding accuracies found across siteswere not due to between-site variability, but rather to the lackof consistent separation across ASD and control individuals.

Classification of Different Group Sizes

We evaluated our classification procedures on smaller, randomlyselected groups of subjects (n = 20, n = 50, n = 100, n = 150, and n =200) to characterize the relationship between group size and thedistribution of decoding accuracies. Randomly selected sub-groups of ASD individuals were age and IQmatched to control in-dividuals and the LDA classification analysis was performedusing the 10-fold and leave-two-out validation schemes (Fig. 7).This procedure was repeated 100 times for each group sizewhile selecting different subjects each time, thereby yielding adistribution of decoding accuracies for each group size. Smallergroup sizes yielded wider decoding accuracy distributions re-gardless of the anatomical measures used to perform the classi-fication. For example, performing this analysis with 148 corticalsurface area measures revealed that ∼30% of the randomly se-lected groups containing 20 subjects in each group (markedwith a star) exhibited decoding accuracies >60%. Note that this ef-fect wasmore prominent in the leave-two-out validation schemeand that the real accuracy computed on the entire ABIDE data inthis case was indistinguishable from chance level (51%). Since

Anatomical Abnormalities in ASD Haar et al. | 5

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

Figure 2. Scatter plots of cortical white matter volume, cortical gray matter volume, cortical thickness, cortical surface area, and cerebellar white matter, cerebellar gray

matter, hippocampus, amygdala, corpus callosum (CC), and ventricular volumes in individualswithASD (gray) andmatched controls (black) separately for each site. Bars:

means across sites for each subject group. Asterisks: significant differences (P < 0.05, randomization test, FDR corrected). Error bars: standard error of the mean across

subjects. d = Cohen’s d (effect size across groups).

6 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

there are multiple feature sets that can be chosen, the use ofsmall groups and the choice of an arbitrary feature set holdsthe potential of greatly inflating decoding accuracies.

Age and Data Sampling Effects

To assess the effects of age on the examined anatomical mea-sures, we also computed growth curves for each of the gross

Figure 3. Scatter plots of corpus callosum segment volumes in individuals with ASD (gray) andmatched controls (black) separately for each site. Bars: means across sites

for each subject group. Asterisks: significant differences (P < 0.05, randomization test, FDR corrected). Error bars: standard error of the mean across subjects. d = Cohen’s d

(effect size across groups).

Figure 4.Cortical ROIs thatwere significantly thicker in the ASD group comparedwith the control group. Therewere no significant differences in the opposite comparison.

Significance was assessed using a t-test and FDR was used to correct for multiple comparisons.

Anatomical Abnormalities in ASD Haar et al. | 7

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

anatomy measures in each of the groups. We used a linear, anexponential, and a quadratic term to fit the data of the ASDand control subjects separately. The use of the 3 predictors en-abled us to avoid making any assumptions regarding the shapeof the fitted curve. The estimated parameters did not differ sig-nificantly across groups (P > 0.3, randomization test). We,

therefore, performed a multiple regression analysis for the en-tire sample (ASD and controls) and regressed out the variabilityassociated with age, effectively flattening the growth curves ofboth groups (Supplementary Fig. 5), and performed the ana-lyses described above using the residuals of this regressionanalysis.

Figure 5. Correlation of intracranial volume, white matter volume, gray matter volume, cortical thickness, and cortical surface area, with IQ in the control (top row) and

autism (middle row) groups or with ADOS (bottom row). Asterisks: significant correlation (P < 0.05, randomization analysis, uncorrected to increase sensitivity).

Figure 6.Decoding accuracies. Classification of group identitywas performed separately for all subjects and threeASD subgroupswhohadADOS scores <10, ADOS = 10–14,

or ADOS >14. Bars show decoding accuracies for each of the following feature sets: all 180 ROI volumes, 32 subcortical ROI volumes, 148 cortical ROI volumes, 148 cortical

ROI thicknesses, and 148 cortical ROI surface areas. Error bars: 5th and 95th percentiles of chance accuracy distribution as estimated with a randomization analysis.

Asterisks: significant decoding accuracies (P < 0.05, uncorrected to increase sensitivity).

8 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

In afinal set of analyses, we demonstrate that similar findingsto those reported above are apparentwhen relaxing inclusion cri-teria so as to include the vast majority of viable MRI scans avail-able in the ABIDE database. Here, we selected 906 subjects (halfwith ASD) without matching ASD and control subjects with re-spect to gender, IQ or age. Our only sampling criterion here wasto ensure that the number of ASD and control subjects wasequal at every site. Equivalent comparisons of the gross anatom-ical measures and ROIs of special interest (after regressing outthe variance associated with site and age) revealed that theASD group exhibited significantly smaller white matter volume,smaller cerebellar gray matter volumes, larger cortical thickness,and larger ventricles volume than the control group (Supplemen-tary Fig. 6). Here too, effect sizes were small (cohen’s d ≤ 0.32 forall measures) with large between-subject variability apparent inboth groups. An equivalent classification analysis using the LDAclassifier and the 10-fold validation scheme showed that decod-ing accuracies were significantly above-chance when using allROI volumes, subcortical volumes, and cortical thickness mea-sures (P < 0.05, randomization analysis, uncorrected), yet all de-coding accuracies were below 60% (Supplementary Fig. 7). Notethat the slightly larger between-group differences apparent inthis larger sample may have been driven by age, IQ, and genderdifferences across ASD and control groups that were not equatedin this sample.

Choice of Segmentation Tool

The analyses above were performed using anatomical measuresthat were computed with the Freesurfer segmentation and par-cellation algorithms. We also performed an additional analysisusing the FAST tissue-type segmentation algorithm, which ispart of the FSL toolbox (Jenkinson et al. 2012). This algorithm islimited in comparison to Freesurfer and enables extraction ofonly 3 anatomical measures: total gray matter (cortical and cere-bellar), white matter (cortical and cerebellar), and CSF volumes.Comparisons of these measures across the 2 groups showed,again, no significant differences across groups and large with-in-group variability (Fig. 8).

DiscussionAnalyses of anatomical MRI scans from the ABIDE databaserevealed several weak yet significant differences across ASDand control groups. Of the 180+ examined ROIs, significant

differences were found only in ventricular volumes, the centralsegment of the CC, and in cortical thickness measures of severaloccipital and temporal ROIs (Figs 2 and 4). Therewas no evidencefor between-group differences in anymeasures of gross anatomyor in specific brain regions including the amygdala, hippocam-pus, most segments of the CC, and the cerebellum, whichhave been implicated in previous anatomical studies of ASD(Figs 1–3). These results suggest that many of the previouslyreported anatomical abnormalities are likely to be of low scientif-ic and clinical significance for explaining ASD neuropathology asa whole in individuals 6–35 years of age. Instead, it may be moreuseful to divide the heterogeneous ASD population into genetic-ally, clinically, and/or behaviorally homogeneous subgroups andattempt to identify unique anatomical abnormalities in each.

Attempts to decode the group identity of single subjects withmultivariate classification techniques using different sets of ana-tomical measures revealed remarkably weak decoding accur-acies. While decoding accuracies using volumetric measures ofsubcortical areas and cortical thickness measures were signifi-cantly above-chance level, all accuracies were below 60% (Fig. 6and Supplementary Figs 3 and 7). Poor decoding accuracies dem-onstrate that weak anatomical differences are apparent only atthe group level and offer very limited diagnostic value at the sin-gle-subject level, likely due to the considerable anatomical vari-ability across subjects of each group.

Taken together, our results suggest that individuals with ASDages 6–35 years old who are capable of participating in MRI stud-ies (i.e., high-functioning individuals with IQ >70), present ana-tomical profiles that are mostly indistinguishable from those ofcontrol individuals. Small differences across groups were greatlyovershadowed by considerablewithin-group variability apparentin both ASD and control groups (Figs 1–3). While consistent ana-tomical abnormalities may be evident in toddlers with ASD dur-ing early development (Courchesne et al. 2004, 2007; Amaral et al.2008; Stanfield et al. 2008), our results suggest that these abnor-malities are not apparent in children, adolescents, and adultswith ASD despite their persisting behavioral problems.

Gross Anatomical Findings in ASD

Numerous studies have reported that toddlers aged 1–4 years oldwho are later diagnosed with ASD exhibit gross anatomical dif-ferences including increased head circumferences and intracra-nial volumes in comparison to control toddlers (Courchesneet al. 2001, 2004, 2011; Amaral et al. 2008; Shen et al. 2013). In con-trast to the studies in toddlers, there is considerable controversyregarding the presence and significance of gross anatomical ab-normalities in individuals with ASD above the age of 6 years.While some studies have reported that individuals with ASD ex-hibit increased gray (Lotspeich et al. 2004; Palmen et al. 2004; Ha-zlett et al. 2006; Ecker et al. 2013) and white (Hazlett et al. 2006)matter volumes, others have reported no group difference(Ecker, Rocha-Rego et al. 2010; Ecker et al. 2012). Additional stud-ies have reported that individuals with ASD exhibit significantlythicker cortices (Hardan et al. 2006) while others have reportedthe opposite (Scheel et al. 2011). The definitive findings reportedhere show that intracranial, gray matter, and white mattervolumes, overall cortical thickness, and surface area measuresare not significantly different across groups (Figs 1 and 2).These results suggest that persistent ASD behavioral symptomsare not ubiquitously associatedwith gross anatomical abnormal-ities in 6–35 year old individuals with ASD, regardless of theirsymptom severity (Supplementary Fig. 2). The transition frompositive to negative findings when examining toddlers versus

Figure 7. Decoding accuracy distributions when performing classification

analyses with random samples of 20 (star), 50 (diamond), 100 (circle), 150

(triangle), and 200 (square) subjects in each group using cortical surface area

measures. Black line: decoding rate for entire ABIDE data with 10-fold (left) and

leave-two-out (right) validation schemes.

Anatomical Abnormalities in ASD Haar et al. | 9

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

older individuals supports recent hypotheses, which have sug-gested that ASD is characterized by early brain overgrowth fol-lowed by arrested growth or even degeneration at later ages(Courchesne et al. 2011).

Brain Areas of Special Interest

Similar controversy exists regarding the presence and signifi-cance of anatomical abnormalities in several ROIs of specificinterest to ASD research including the amygdala, hippocampus,cerebellum, and CC. The “Amygdala theory of autism” has sug-gested that abnormalities in the amygdala may explain the so-cial and emotional difficulties found in ASD (Baron-cohen et al.2000). While some studies have reported significantly largeramygdala volumes in ASD individuals than controls, othershave reported no difference across groups, and one study hasreported the opposite (Bellani et al. 2013a). There were nosignificant differences across groups in our study in amygdalavolumes (Fig. 2) nor in hippocampus volumes, another limbicsystem region of particular interest for ASD research (Sweetenet al. 2002).

The CC has also been a focus of attention following severaltheories that have proposed poor long-range neural connectiv-ity/synchronization in ASD (Gepner and Féron 2009). Whilesome studies have reported significantly smaller CC volumes inASD individuals than controls, others have reported no differ-ence across groups (Bellani et al. 2013b). We found no significantbetween-group difference when assessing the entire CC volume(Fig. 2), yet a more detailed analysis revealed that the ASDgroup exhibited a significantly smaller volume in the central seg-ment of the CC when compared with the control group (Fig. 3).The effect size of this difference, however, was very small(Cohen’s d = 0.2) and stands in contrast to previous estimates ofmuch larger effect sizes in all CC segments (Cohen’s d > 0.4,Frazier and Hardan 2009).

Studies of the cerebellum in ASD individuals have reportedmixed findings, as well, with some documenting abnormally

large cerebellar volumes (Hardan, Minshew, Harenski et al.2001; Scott et al. 2009) and abnormally small vermis volumesandmidsagittal vermis areas, often found only in specific lobules(Courchesne et al. 1988, 1994). Here, we only assessed cerebellargray and white matter volumes and found no significant differ-ences across groups (Fig. 2).

Ventricles

While early ventriculomegaly (enlarged ventricles) is associatedwith multiple developmental delays (McKechnie et al. 2012),only a few studies have assessed ventricle volumes in ASD withsome (Palmen et al. 2004) but not others (Hardan, Minshew,Mallikarjuhn et al. 2001) reporting significantly larger ventriclevolumes in ASD. Our results revealed significantly larger ven-tricle volumes in ASD when compared with controls (Fig. 2), yetdifferences across groups exhibited a relatively small effect size(Cohen’s d = 0.34).

Relationship Between Volume, IQ, and ADOS

Our findings are consistent with previous studies that have re-ported significant correlations between IQ and gross anatomicalmeasures (Deary et al. 2010). Both control and ASD subjects ex-hibited significant positive correlations between IQ and intracra-nial, gray, and white matter volumes as well as cortical surfacearea (Fig. 5). ASD individuals also exhibited significant positivecorrelations between ADOS scores and gray and white mattervolumes as well as cortical surface area. Considering that grossanatomical measures are a function of neuronal loss, synapticpruning, and gray matter shrinkage, in conjunction with whitematter maturation and growth (as evident in the growth curvesof Supplementary Fig. 5), it is intriguing that similar positivecorrelations exist between gross anatomical developmentalmeasures and IQ/ADOS scores. Our results, however, offervery limited insight regarding the meaning of these generalassociations.

Figure 8. Results from the FSL FAST algorithm revealed no significant differences across groups. Scatter plots of cerebrospinal fluid (CSF), gray matter, and white matter

volumes in individuals with ASD (gray) and matched controls (black) separately for each site. Bars: means across sites for each subject group. Asterisks: significant

differences (P < 0.05, randomization test, FDR corrected). Error bars: standard error of the mean across subjects. d = Cohen’s d (effect size across groups).

10 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

When considering the IQ range of the current ASD sample, itis important to note that our results are limited to relatively high-functioning ASD individuals (IQ mean = 106, standard deviation,SD = 17). ASD individuals with lower functionmay potentially ex-hibit significant structural abnormalities and stronger correla-tions with behavioral measures. Studies with ASD individualsof lower function are generally missing from the ASD neuroima-ging literature and are highly warranted.

Multivariate Classification of Autism

Multivariate classification techniques offer a method for decod-ing the group identity of ASD and control subjects based on pat-terns of anatomical features, rather than single anatomicalmeasures. Several recent studies have reported exceptional de-coding accuracies (over 85%) of group identity when using mea-sures of cortical thickness, geometry, curvature, and/or surfacearea in small samples of 20–30 subjects from each group (Ecker,Marquand et al. 2010; Jiao et al. 2010; Uddin et al. 2011). Thesestudies have suggested that distributed anatomical abnormal-ities can consistently be used to identify adults with ASD.

Performing similar multivariate analyses with the ABIDE datarevealed much lower decoding accuracies (<60%), whichexceeded chance levels only when usingmeasures of subcorticalvolumes or cortical thickness (Fig. 6). Similarly weak decodingaccuracies were found when using different validation schemes(i.e., leave-two-out instead of 10-fold validation) and whenusing linear or nonlinear classification algorithms (Supplemen-tary Fig. 3). In addition, performing the same classification ana-lyses within each of the 5 largest sites (>20 subjects in eachgroup) revealed equivalent results with accuracy rates below60% in all sites. These results suggest that weak classification ac-curacies were not generated by differences across sites or thechoice of a particular classification algorithm or validationscheme. Instead, we believe that theseweak decoding accuracieswere the consequence of the considerable within-group variabil-ity, which was clearly evident in the univariate analyses (Figs 1–3). Variable and inconsistent anatomy across individuals of eachgroup is expected to reduce classification accuracy across groups.

While classification techniques hold great promise, they alsosuffer fromamajor potential pitfall: overfitting. The accuracyandvalidity of a classifier depend on the number of features beingconsidered and the number of subjects available for trainingand testing. In situations where there are more anatomical fea-tures than subjects, the classification algorithm may arbitrarilyseparate individual subjects into meaningless groups by overfit-ting noise/variability (Cawley and Talbot 2010). When only asmall number of subjects are left out for testing, there is a riskof arbitrarily inflating decoding accuracies. This was clearly ap-parent when we performed classification analyses with small,randomly selected subgroups. For example, when decoding sub-ject identity based on cortical thickness measures, the actual de-coding accuracy was 51%, yet roughly 30% of randomly selectedsmall groups (20 ASD/20 control subjects) exhibited decoding ac-curacies above 60%. When considering the multitude of anatom-ical features that can be used for classification, the potential offinding a set of features that yields accurate decoding seems ex-tremely high. It is, therefore, critical to evaluate classificationschemes on large samples and test their generalizability.

Automated Analyses Using Large Multicenter Samples

We believe that the controversy regarding the presence and sig-nificance of anatomical abnormalities in ASD is a result of the

fact that the majority of previous studies recruited small ASDand control samples with distinct age, IQ, and ASD severity char-acteristics from general populations that have considerable ana-tomical variability. We suggest that studies with such smallsamples are unlikely to capture the true anatomical distributionsof either population and are, therefore, unlikely to yield consist-ent results. The only way to identify robust differences acrossASD and control groups and reconcile previous contradictionsin the literature is to assess anatomical differences across largecohorts where age, IQ, gender, and autism severity factors canbe controlled for (or their relative contribution assessed) andthe true anatomical distribution of each measure can be esti-mated for each group as performed here using the ABIDE data.

The down side of evaluating the ABIDE data is that aggregatedMRI scans from multiple international sites may contain largebetween-site variability, resulting from possible confoundingfactors, for example, autism diagnosis, and hence, subject selec-tion may differ across sites such that sites with unique clinicalcriteria may “contaminate” the dataset. All of the participantsin the current study, however, were characterized with the stan-dardized ADOS assessment, which is the current “golden stand-ard” for identifying individuals with ASD. Another possibleconfound is the heterogeneity ofMRI scanners and scanning pro-tocols across sites. Fortunately, however, automated estimationsof cortical and subcortical volumes with Freesurfer have beenshown to be remarkably robust across MRI scanners of differentmanufacturers (varying by 0–3%) as long as the field strength iskept constant at 3T (Jovicich et al. 2009). The third and likely lar-gest source of variability across sites is generated by genetic andenvironmental heterogeneity, which are known to affect brainanatomy (Thompson et al. 2001).

To address this caveat, we performed several analyses to en-sure that our results were not due to between-site variability.First, we used a standard mixed-model approach to examinethe contribution of site differences to potential differences be-tween ASD and control groups. In an additional step, we re-gressed out the differences across sites and performed severalunivariate and multivariate analyses with data where the meanof each site had been normalized. Finally, we performed per-siteanalyses to determine reliability of results across multiple sites.These analyses reassured us that the conclusions of this studywere not based on potentially misleading between-site differ-ences or on spurious findings that were apparent in only asmall minority of sites. Perhaps the strongest impression oneshould gain from examining the presented data is that the vastmajority of sites do not show any consistent anatomical differ-ences across groups. Importantly, in order for group differencesto be clinically relevant, they must be robust to differences inMRI scanning parameters as well as genetic and environmentalheterogeneity.

Implications for Future ASD Research

The current study demonstrates that anatomical differences be-tween high-functioning ASD and control groups (aged 6–35 yearsold) are very small in comparison to large within-group variabil-ity. This suggests that anatomical measures alone are likely to beof low scientific and clinical significance for identifying children,adolescents, and adults with ASD or for elucidating their neuro-pathology. Rather than expecting to find consistent anatomicalabnormalities across the entire ASD population, it may be morereasonable to search for subgroups of ASD individuals withmore homogeneous etiologies who may (or may not) exhibitcommon structural findings. Determining how to segregate

Anatomical Abnormalities in ASD Haar et al. | 11

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

ASD individuals into meaningful subgroups based on geneticprofiling, clinical comorbidities, sensory sensitivities, and otherrelevant measures seems to be the most urgent next step for fu-ture ASD research. Further efforts to aggregate broad clinical,genetic, andneuroimaging data from largeASD samples of differ-ent ages and IQ levels will be extremely helpful in achieving thisgoal.

Supplementary MaterialSupplementary material can be found at: http://www.cercor.ox-fordjournals.org/.

FundingThis work was supported by Simons Foundation SFARI grant177638 (M.B. and I.D.).

NotesWe thank our colleague, Dr Nancy Minshew, for her consistentsupport of our research efforts. We also thank Dr Joel Green-house, Department of Statistics, Carnegie Mellon University, forhis useful guidance on some of the analyses. Conflict of Interest:None declared .

ReferencesAmaral DG, Schumann CM, Nordahl CW. 2008. Neuroanatomy of

autism. Trends Neurosci. 31:137–145.Baron-cohen S, Ring HA, Bullmore ET, Wheelwright S, Ashwin C,

Williams SC. 2000. The amygdala theory of autism. NeurosciBiobehav Rev. 24:355–364.

Bellani M, Calderoni S, Muratori F, Brambilla P. 2013a. Brain anat-omy of autism spectrum disorders II. Focus on amygdala.Epidemiol Psychiatr Sci. 22:309–312.

Bellani M, Calderoni S, Muratori F, Brambilla P. 2013b. Brain anat-omy of autism spectrum disorders I. Focus on corpus callo-sum. Epidemiol Psychiatr Sci. 22:217–221.

Benjamini Y, Hechtlinger Y. 2014. Discussion: an estimate of thescience-wise false discovery rate and applications to topmed-ical journals by Jager and Leek. Biostatistics. 15:13–16; discus-sion 39–45.

Cawley G, Talbot N. 2010. On over-fitting in model selection andsubsequent selection bias in performance evaluation. J MachLearn Res. 11:2079–2107.

Courchesne E, Campbell K, Solso S. 2011. Brain growth across thelife span in autism: age-specific changes in anatomical path-ology. Brain Res. 1380:138–145.

Courchesne E, Karns CM, Davis HR, Ziccardi R, Carper RA,Tigue ZD, ChisumHJ, Moses P, Pierce K, Lord C, et al. 2001. Un-usual brain growth patterns in early life in patients with aut-istic disorder. An MRI study. Neurology. 57:245–254.

Courchesne E, Pierce K, Schumann CM, Redcay E, Buckwalter Ja,Kennedy DP, Morgan J. 2007. Mapping early brain develop-ment in autism. Neuron. 56:399–413.

Courchesne E, Redcay E, Kennedy DP. 2004. The autistic brain:birth through adulthood. Curr Opin Neurol. 17:489–496.

Courchesne E, Townsend J, Saitoh O. 1994. The brain in infantileautism: posterior fossa structures are abnormal. Neurology.44:214–223.

Courchesne E, Yeung-Courchesne R, Press GA, Hesselink JR,Jernigan TL. 1988. Hypoplasia of cerebellar vermal lobules VIand VII in autism. N Engl J Med. 318:1349–1354.

Deary IJ, Penke L, Johnson W. 2010. The neuroscience of humanintelligence differences. Nat Rev Neurosci. 11:201–211.

Destrieux C, Fischl B, Dale A, Halgren E. 2010. Automatic parcella-tion of human cortical gyri and sulci using standard anatom-ical nomenclature. Neuroimage. 53:1–15.

Di Martino A, Yan C-G, Li Q, Denio E, Castellanos FX, Alaerts K,Anderson JS, Assaf M, Bookheimer SY, Dapretto M, et al.2014. The autism brain imaging data exchange: towards alarge-scale evaluation of the intrinsic brain architecture inautism. Mol Psychiatry. 19:659–667.

Ecker C, Ginestet C, Feng Y, Johnston P, Lombardo MV, Lai M-C,Suckling J, Palaniyappan L, Daly E, Murphy CM, et al. 2013.Brain surface anatomy in adults with autism: the relationshipbetween surface area, cortical thickness, and autistic symp-toms. JAMA Psychiatry. 70:59–70.

Ecker C, Marquand A, Mourão-Miranda J, Johnston P, Daly EM,Brammer MJ, Maltezos S, Murphy CM, Robertson D,Williams SC, et al. 2010. Describing the brain in autism infive dimensions—magnetic resonance imaging-assisted diag-nosis of autism spectrum disorder using a multiparameterclassification approach. J Neurosci. 30:10612–10623.

Ecker C, Rocha-Rego V, Johnston P, Mourao-Miranda J,Marquand A, Daly EM, Brammer MJ, Murphy C, Murphy DG.2010. Investigating the predictive value of whole-brain struc-tural MR scans in autism: a pattern classification approach.Neuroimage. 49:44–56.

Ecker C, Suckling J, Deoni SC, Lombardo MV, Bullmore ET, Baron-Cohen S, Catani M, Jezzard P, Barnes A, Bailey AJ, et al. 2012.Brain anatomy and its relationship to behavior in adultswith autism spectrum disorder: a multicenter magnetic res-onance imaging study. Arch Gen Psychiatry. 69:195–209.

Fatemi SH, Aldinger KA, Ashwood P, Bauman ML, Blaha CD,Blatt GJ, Chauhan A, Chauhan V, Dager SR, Dickson PE, et al.2012. Consensus paper: pathological role of the cerebellumin autism. Cerebellum. 11:777–807.

Fischl B. 2012. FreeSurfer. Neuroimage. 62:774–781.Fischl B, Dale AM. 2000. Measuring the thickness of the human

cerebral cortex from magnetic resonance images. Proc NatlAcad Sci USA. 97:11050–11055.

Frazier TW, Hardan AY. 2009. A meta-analysis of the corpus cal-losum in autism. Biol Psychiatry. 66:935–941.

Gepner B, Féron F. 2009. Autism: a world changing too fast for amis-wired brain? Neurosci Biobehav Rev. 33:1227–1242.

Groen W, Teluij M, Buitelaar J, Tendolkar I. 2010. Amygdala andhippocampus enlargement during adolescence in autism.J Am Acad Child Adolesc Psychiatry. 49:552–560.

Hardan AY, Minshew NJ, Harenski K, Keshavan MS. 2001. Poster-ior fossa magnetic resonance imaging in autism. J Am AcadChild Adolesc Psychiatry. 40:666–672.

Hardan AY, Minshew NJ, Mallikarjuhn M, Keshavan MS. 2001.Brain volume in autism. J Child Neurol. 16:421–424.

Hardan AY, Muddasani S, Vemulapalli M, Keshavan MS,Minshew NJ. 2006. An MRI study of increased cortical thick-ness in autism. Am J Psychiatry. 163:1290–1292.

Hazlett HC, Poe MD, Gerig G, Smith RG, Piven J. 2006. Cortical grayandwhite brain tissue volume in adolescents and adults withautism. Biol Psychiatry. 59:1–6.

Jenkinson M, Beckmann CF, Behrens TEJ, Woolrich MW,Smith SM. 2012. FSL. Neuroimage. 62:782–790.

Jiao Y, Chen R, Ke X, Chu K, Lu Z, Herskovits EH. 2010. Predictivemodels of autism spectrum disorder based on brain regionalcortical thickness. Neuroimage. 50:589–599.

Jovicich J, Czanner S, Han X, Salat D, van der Kouwe A, Quinn B,Pacheco J, Albert M, Killiany R, Blacker D, et al. 2009. MRI-

12 | Cerebral Cortex

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from

derivedmeasurements of human subcortical, ventricular andintracranial brain volumes: reliability effects of scan sessions,acquisition sequences, data analyses, scanner upgrade, scan-ner vendors and field strengths. Neuroimage. 46:177–192.

JustMA,CherkasskyVL,KellerTA,KanaRK,MinshewNJ. 2007. Func-tional and anatomical cortical underconnectivity in autism: evi-dence from an fMRI study of an executive function task andcorpus callosum morphometry. Cereb Cortex. 17:951–961.

Lotspeich LJ, Kwon H, Schumann CM, Fryer SL, Goodlin-Jones BL,Buonocore MH, Lammers CR, Amaral DG, Reiss AL. 2004. In-vestigation of neuroanatomical differences between autismand Asperger syndrome. Arch Gen Psychiatry. 61:291–298.

McKechnie L, Vasudevan C, LeveneM. 2012. Neonatal outcome ofcongenital ventriculomegaly. Semin Fetal Neonatal Med.17:301–307.

Palmen SJMC, Hulshoff Pol HE, Kemner C, Schnack HG, Janssen J,Kahn RS, van Engeland H. 2004. Larger brains in medicationnaive high-functioning subjects with pervasive developmen-tal disorder. J Autism Dev Disord. 34:603–613.

Raznahan A, Toro R, Daly E, Robertson D, Murphy C, Deeley Q,Bolton PF, Paus T,MurphyDGM. 2010. Cortical anatomy in aut-ism spectrum disorder: an in vivo MRI study on the effect ofage. Cereb Cortex. 20:1332–1340.

Scheel C, Rotarska-Jagiela A, Schilbach L, Lehnhardt FG, Krug B,Vogeley K, Tepest R. 2011. Imaging derived cortical thicknessreduction in high-functioning autism: key regions and tem-poral slope. Neuroimage. 58:391–400.

Scott JA, Schumann CM, Goodlin-Jones BL, Amaral DG. 2009. Acomprehensive volumetric analysis of the cerebellum in chil-dren and adolescents with autism spectrum disorder. AutismRes. 2:246–257.

Shen MD, Nordahl CW, Young GS, Wootton-Gorges SL, Lee A,Liston SE, Harrington KR, Ozonoff S, Amaral DG. 2013.Early brain enlargement and elevated extra-axial fluid in in-fants who develop autism spectrum disorder. Brain. 136:2825–2835.

Stanfield AC, McIntosh AM, Spencer MD, Philip R, Gaur S,Lawrie SM. 2008. Towards a neuroanatomy of autism: a sys-tematic review and meta-analysis of structural magnetic res-onance imaging studies. Eur Psychiatry. 23:289–299.

Sweeten TL, Posey DJ, Shekhar A, McDougle CJ. 2002. The amyg-dala and related structures in the pathophysiology of autism.Pharmacol Biochem Behav. 71:449–455.

Thompson PM, Cannon TD, Narr KL, van Erp T, Poutanen VP,Huttunen M, Lönnqvist J, Standertskjöld-Nordenstam CG,Kaprio J, Khaledy M, et al. 2001. Genetic influences on brainstructure. Nat Neurosci. 4:1253–1258.

Uddin LQ, Menon V, Young CB, Ryali S, Chen T, Khouzam A,MinshewNJ, Hardan AY. 2011. Multivariate searchlight classi-fication of structural magnetic resonance imaging in childrenand adolescents with autism. Biol Psychiatry. 70:833–841.

Wallace GL, Dankner N, Kenworthy L, Giedd JN, Martin A. 2010.Age-related temporal and parietal cortical thinning in autismspectrum disorders. Brain. 133:3745–3754.

Anatomical Abnormalities in ASD Haar et al. | 13

at Acquisitions D

eptHunt L

ibrary on October 15, 2014

http://cercor.oxfordjournals.org/D

ownloaded from